Updating Rules for Non-Bayesian Preferences. Tan Wang∗ December, 1999 Abstract We axiomatize updating rules for preferences that are not necessarily in the expected utility class. Two sets of results are presented. The first is the axiomatization and representation of conditional preferences. The second consists of the axiomatization of three updating rules: the traditional Bayesian rule, the Dempster-Shafer rule, and the generalized Bayesian rule. The last rule can be regarded as the updating rule for the multi-prior expected utility (Gilboa and Schmeidler (1989)). The operational merit of it is that it is equivalent to updating each prior by the traditional Bayesian rule. ∗Tan Wang is with the Faculty of Commerce and Business Administration, University of British Columbia, Van- couver, British Columbia, Canada, V6T 1Z2.
[email protected]. The author is grateful to the Social Sciences and Humanities Research Council of Canada for financial support. 1 Introduction The traditional approach to updating is the Bayesian rule. This approach is justified by the ax- iomatic treatment of Savage (1954), where it is shown that, in situations of uncertainty, if a decision maker’s preference satisfies a certain set of axioms, his preference can be represented by an expected utility with respect to a subjective probability measure and that probability measure represents the decision maker’s belief about the likelihood of events. Moreover, in light of new information, the decision maker updates his belief according the Bayesian rule. This Savage paradigm has been the foundation of much of the economic theories under uncertainty. At the same time, however, the Savage paradigm has been challenged by behavior exhibited in Ellsberg paradox (Ellsberg (1961)), which seems to question the very notion of representing a decision maker’s belief by a probability measure and hence by implication the validity of the Bayesian rule.